Method and device for fast prediction of three-dimensional nozzle flow field and sensitivity parameter analysis

Abstract

The application provides a three-dimensional nozzle flow field rapid prediction and sensitive parameter analysis method and device, wherein the method comprises the following steps: obtaining nozzle geometric shape parameters, performing feature extraction on the geometric shape parameters by using a self-encoder to obtain geometric feature parameters; reconstructing the nozzle geometric shape parameters through the geometric feature parameters, taking nozzle flow state parameters and flow control parameters as design variables, and obtaining nozzle flow field training data sets through CFD simulation; building a multilayer perceptron neural network, training the multilayer perceptron neural network through the training data sets, introducing a flow control equation into a loss function of the training, and obtaining a flow field prediction model; and realizing three-dimensional flow field rapid prediction of specified nozzle geometric feature parameters, flow state parameters and flow control parameters through the flow field prediction model to obtain flow field information of the specified nozzle. The application realizes high-precision and rapid prediction of three-dimensional flow field distribution and completes sensitive parameter analysis of the nozzle.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of aircraft power machinery and artificial intelligence application. BACKGROUND

[0002] Current flow field calculation for a nozzle mainly adopts the method of characteristic (MOC), CFD and various existing proxy model methods of flow field calculation. The method of characteristic has small calculation amount and fast solving speed but low precision, and is mainly used for calculation in the early stage of design. The CFD method has large calculation amount and high solving precision but slow speed, so the high-precision aerodynamic calculation method based on CFD is time-consuming. The proxy model method usually has high calculation efficiency, but the solving target is mainly the characteristic quantity such as force coefficient corresponding to the flow field, and the solving precision is determined according to the specific method. The flow field prediction proxy model based on the CFD solving result can theoretically realize the same function as the CFD calculation and has the same order of solving precision, but the existing methods all have their own defects. Binjian Zhang et al. use a radial basis neural network to construct a CFD proxy model for a two-dimensional airfoil flow problem, directly associate the airfoil geometric parameters with the corresponding aerodynamic characteristic parameters, and have fast solving speed and high precision, but the neural network layer of the proxy model constructed by them is shallow, the calculation ability is limited, the flow parameters such as velocity and pressure distribution in the whole flow field cannot be obtained, and therefore the CFD method cannot be replaced. SUMMARY

[0003] The present application aims to at least partially solve one of the problems in the related art.

[0004] To this end, the first object of the present application is to propose a three-dimensional nozzle flow field fast prediction and sensitivity parameter analysis method for realizing high-precision fast prediction of three-dimensional flow field distribution and completing sensitivity parameter analysis of the nozzle.

[0005] The second object of the present application is to propose a three-dimensional nozzle flow field fast prediction and sensitivity parameter analysis device.

[0006] The third object of the present application is to propose a computer device.

[0007] The fourth object of the present application is to propose a computer readable storage medium.

[0008] To achieve the above object, the embodiment of the first aspect of the present application provides a three-dimensional nozzle flow field fast prediction and sensitivity parameter analysis method, comprising: obtaining nozzle geometric shape parameters, using a self-encoder neural network to extract features of the geometric shape parameters to obtain geometric feature parameters; reconstructing the nozzle geometric shape parameters through the geometric feature parameters, taking nozzle flow state parameters and flow control parameters as design variables, and obtaining nozzle flow field training data set through CFD simulation; building a multilayer perceptron neural network, training the multilayer perceptron neural network through the training data set, and introducing a flow control equation into a loss function of the training to obtain a flow field prediction model; and realizing three-dimensional flow field fast prediction of specified nozzle geometric feature parameters, flow state parameters and flow control parameters through the flow field prediction model to obtain flow field information of the specified nozzle.

[0009] The three-dimensional nozzle flow field fast prediction and sensitivity parameter analysis method provided by the embodiment of the present application fully combines the self-encoder and the multilayer perceptron technology with embedded physical constraints to build a three-dimensional flow field fast prediction deep neural network for the nozzle; through parameterized modeling, nozzle geometric data set construction is realized; through writing grid division software and corresponding script files of a computational fluid dynamics solver, large-scale batch simulation is realized to build a flow field data set required for neural network training; by changing the nozzle geometric parameters and the flow control parameters within a certain range and giving the flow state parameters, the trained flow field fast prediction deep neural network is called to obtain the corresponding flow field velocity, pressure and other parameter distributions in real time, and the characteristic physical quantities such as nozzle lift, thrust and thrust offset under various working conditions are obtained through post-processing, and then the sensitivity parameters of the nozzle are obtained by using a statistical method; the flow field pressure gradient is calculated to directly observe the flow field wave distribution and analyze the action mechanism of the sensitivity parameters. The solution method provided by the present application has a wide application range, greatly improves the three-dimensional flow field calculation efficiency of the nozzle, ensures the flow field prediction accuracy to be the same order of magnitude as the CFD simulation, and reduces the time required for sensitivity analysis.

[0010] In addition, the three-dimensional nozzle flow field fast prediction and sensitivity parameter analysis method according to the above embodiment of the present application can also have the following additional technical features:

[0011] Further, in an embodiment of the present application, the self-encoder neural network is used to extract features of the geometric shape parameters to obtain geometric feature parameters, and further comprises:

[0012] Random noise is added to the geometric shape parameters.

[0013] Further, in an embodiment of the present application, before the CFD simulation, further comprising:

[0014] The grid division and boundary conditions are set.

[0015] Further, in an embodiment of the present application, after the three-dimensional flow field is quickly predicted for the specified nozzle geometric feature parameters, flow state parameters and flow control parameters through the flow field prediction model, the flow field information is obtained, and the method further comprises:

[0016] The aerodynamic characteristic parameters of the specified nozzle are obtained through the three-dimensional flow field information, including lift, thrust, core volume and thrust bias degree.

[0017] Further, in an embodiment of the present application, the method further comprises:

[0018] The aerodynamic characteristics of the nozzle under different flow state parameters and flow control parameters are solved by changing the geometric feature parameters of the specified nozzle and using the flow field prediction model, so as to screen out the sensitive parameters.

[0019] Further, in an embodiment of the present application, the three-dimensional flow field information of the specified nozzle is obtained by quickly predicting the three-dimensional flow field for the specified nozzle geometric feature parameters, flow state parameters and flow control parameters through the flow field prediction model, and the method further comprises:

[0020] The flow field pressure gradient is obtained through the three-dimensional flow field information, and the mechanism of the sensitive parameters is analyzed through the wave distribution.

[0021] To achieve the above purpose, a second aspect of the present application provides a three-dimensional nozzle flow field quick prediction and sensitive parameter analysis device, comprising: an extraction module for obtaining nozzle geometric shape parameters, using an autoencoder neural network to extract features of the geometric shape parameters to obtain geometric feature parameters; a construction module for reconstructing the nozzle geometric shape parameters through the geometric feature parameters, taking the nozzle flow state parameters and flow control parameters as design variables, and obtaining the flow field training data set of the nozzle through CFD simulation; a training module for building a multilayer perceptron neural network, training the multilayer perceptron neural network through the training data set, and introducing the flow control equation into the loss function of the training to obtain a flow field prediction model; a prediction module for quickly predicting the three-dimensional flow field for the specified nozzle geometric feature parameters, flow state parameters and flow control parameters through the flow field prediction model, and obtaining the flow field information of the specified nozzle.

[0022] Further, in an embodiment of the present application, the method further comprises an analysis module for:

[0023] The aerodynamic characteristic parameters of the specified nozzle are obtained through the three-dimensional flow field information, including lift, thrust, core volume and thrust bias degree.

[0024] The aerodynamic characteristics of the nozzle under different flow state parameters and flow control parameters are solved by changing the geometric feature parameters of the specified nozzle and using the flow field prediction model, so as to screen out the sensitive parameters.

[0025] The flow field pressure gradient is obtained through the three-dimensional flow field information, and the action mechanism of the sensitive parameter is analyzed through the wave series distribution.

[0026] To achieve the above object, the third aspect of the present application provides a computer device, characterized by comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to realize the three-dimensional nozzle flow field fast prediction and sensitive parameter analysis method as described above.

[0027] To achieve the above object, the fourth aspect of the present application provides a computer readable storage medium, which stores a computer program, characterized in that the computer program is executed by a processor to realize the three-dimensional nozzle flow field fast prediction and sensitive parameter analysis method as described above. BRIEF DESCRIPTION OF DRAWINGS

[0028] The above and / or additional aspects and advantages of the present application will become apparent and more readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings, in which:

[0029] Figure 1 A flow chart of a three-dimensional nozzle flow field fast prediction and sensitive parameter analysis method provided by the embodiments of the present application.

[0030] Figure 2 A flow chart of a three-dimensional nozzle flow field fast prediction and sensitive parameter analysis device provided by the embodiments of the present application.

[0031] Figure 3 A three-dimensional nozzle flow field fast prediction and sensitive parameter analysis flow chart provided by the embodiments of the present application.

[0032] Figure 4 A schematic diagram of an autoencoder principle provided by the embodiments of the present application.

[0033] Figure 5 An AutoEncoder with random noise added provided by the embodiments of the present application.

[0034] Figure 6 A flow chart of extracting geometric feature parameters by an autoencoder provided by the embodiments of the present application.

[0035] Figure 7 A multi-layer perceptron principle diagram provided by the embodiments of the present application.

[0036] Figure 8 A flow chart of constructing a flow field training data set provided by the embodiments of the present application.

[0037] Figure 9A flowchart of a method for predicting flow field distribution based on multilayer perceptron technology provided by an embodiment of the present application.

[0038] Figure 10 An algorithm schematic diagram of embedding physical constraints provided by an embodiment of the present application.

[0039] Figure 11 A flowchart of a method for predicting flow field distribution based on multilayer perceptron technology embedding physical constraints provided by an embodiment of the present application. DETAILED DESCRIPTION

[0040] Embodiments of the present application are described in detail below with reference to examples shown in the attached drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by reference to the drawings are exemplary and are intended to explain the present application, and cannot be understood as limiting the present application.

[0041] A three-dimensional nozzle flow field rapid prediction and sensitivity parameter analysis method and device of an embodiment of the present application are described below with reference to the drawings.

[0042] Figure 1 A flowchart of a three-dimensional nozzle flow field rapid prediction and sensitivity parameter analysis method provided by an embodiment of the present application.

[0043] As shown in Figure 1 the three-dimensional nozzle flow field rapid prediction and sensitivity parameter analysis method includes the following steps:

[0044] S1: Obtain nozzle geometric shape parameters, and use a self-encoder neural network to extract features of the geometric shape parameters to obtain geometric feature parameters;

[0045] Further, in an embodiment of the present application, the self-encoder neural network is used to extract features of the geometric shape parameters to obtain geometric feature parameters, and further includes:

[0046] Random noise is added to the geometric shape parameters.

[0047] As a kind of model in artificial neural network, AutoEncoder adopts unsupervised learning to effectively extract features from high-dimensional data, and has been well applied in academic and industrial circles. Its overall framework includes two modules: encoding process and decoding process. The encoding process is to map the input sample data X from the physical space to the feature space and obtain the high-dimensional feature h. The decoding process is to map the abstract feature h back to the original physical space and reconstruct the sample X'. In the process of training such neural network, the optimization goal is to minimize the reconstruction error (X-X') for the training data set. In the solution, the parameters related to encoding and decoding processes need to be optimized at the same time, so that the neural network can accurately obtain the high-dimensional feature h of the input sample. The simplified flow chart is shown in Figure 4

[0048] As can be seen from the above process, AutoEncoder does not need to use sample labels in the data training process. Essentially, the input of the sample is regarded as the input and output of the neural network at the same time. This unsupervised learning method greatly improves the universality of the method. To further improve the learning accuracy of AutoEncoder and alleviate the problem of easy overfitting in the process of neural network training, random noise can be added to the input sample, as shown in Figure 5 , so as to enhance the robustness of the model and enable the encoding process to learn abstract features with anti-interference ability.

[0049] Therefore, in this project, the AutoEncoder method can be used to complete the extraction of geometric parameter features and reduce the number of parameters. The detailed process is shown in Figure 6 .

[0050] To implement this method, a large number of geometric data for three-dimensional nozzles need to be artificially constructed, and the corresponding training data set is formed, so that the AutoEncoder neural network that can be directly used to accurately extract the geometric feature parameters of the nozzle can be obtained.

[0051] S2: Reconstruct the nozzle geometric shape parameters by geometric feature parameters, take the nozzle flow state parameters and flow control parameters as design variables, and obtain the flow field training data set of the nozzle through CFD simulation;

[0052] Further, in an embodiment of the present application, before the CFD simulation, it further includes:

[0053] Setting grid division and boundary conditions.

[0054] S3: Building a multilayer perceptron neural network, training the multilayer perceptron neural network through the training data set, and introducing the flow control equation into the loss function of the training to obtain a flow field prediction model;

[0055] ​MLP is a kind of forward structure artificial neural network, and its basic structure and calculation process are shown in Figure 7 The final input vector set can be mapped to an output vector set. At least one hidden layer should be contained in this type of model, and each layer is fully connected to the next layer, and each node in the neural network is a neuron with a nonlinear activation function except the input node. The method passes the product of input elements and weights to the summation node with neuron bias, and its main advantage is the ability to quickly solve complex problems, overcoming the defects of perceptron in identifying linearly inseparable data, so it is widely used.

[0056] As a kind of supervised learning algorithm, MLP can continuously train the neural network built by training data set: based on the chain rule of derivation, the error back propagation method is used to calculate the gradient of neural network parameters, so as to update the related parameters such as connection weight and bias, and minimize the overall error between output prediction result and training data.

[0057] In the present application, the multilayer perceptron technology is used to associate the geometric characteristic parameters of three-dimensional nozzle, flow control parameters, flow state parameters and the corresponding flow field parameter distribution under the working condition, so as to realize the rapid acquisition of flow field state (without introducing physical constraints). In order to complete this part of the function, a large number of different nozzle configurations under various flow states are required to train the MLP deep neural network with the corresponding CFD flow field calculation results, so the present project needs to construct the corresponding flow field training data set, and the data set needs to consider the changes of three parameters: flow state parameters, flow control parameters and nozzle geometric characteristic parameters, and the specific process is shown in Figure 8 .

[0058] The solving process mainly includes three key steps: Decoder, parameterized modeling and CFD solving. The Decoder neural network here is the decoding part of AutoEncoder in Figure 6 , and CFD solving can call mainstream CFD commercial software or self-developed high-precision, high-resolution, large-scale parallel CFD program for batch solving. Parameterized modeling needs to realize the construction of the corresponding structural surface of the nozzle by using the complete geometric parameters of the nozzle. The present application adopts polynomial to describe the nozzle profile and other related curves, so the parameterized modeling part is easy to realize. The grid division needs to be completed before CFD solving, and since a large number of different geometric structures are included in the process of building data set, there is a lot of repetitive work, so the related grid processing software, such as Glyph script file in Pointwise, can be used to batch divide and process the grid of different geometric structures designed, and then Batch file (Windows system) is written to call CFD solver in batch, so as to realize the automatic generation of large-scale three-dimensional flow field data set.

[0059] When the data training of the MLP neural network is completed, the CFD proxy model with consideration of accuracy and efficiency can be obtained, as shown in the formula (1). Figure 9 The subsequent research calculation amount is greatly reduced.

[0060] Through relevant research, in order to further improve the flow field prediction accuracy of the MLP model, the present application introduces a physical constraint to realize the MLP neural network embedded with a physical model, namely PI-MLP, so as to accurately capture flow field details such as complex wave distribution of the nozzle under a non-design state.

[0061] Although the deep learning method is used to build and train the neural network, the flow field prediction value under the corresponding input condition can be obtained, but the result cannot guarantee to completely meet the physical flow mechanism, that is, to meet the flow control equation set, and is only a superficial prediction. Therefore, in order to make the result of the flow field prediction model meet the physical law, the flow control equation needs to be introduced into the loss function of the neural network training as an additional constraint term, so that the flow field parameter distribution output by the model is more in line with the physical law.

[0062] The loss function used in the training of the classical MLP deep neural network is as follows:

[0063]

[0064] In the formula, U=(u, v, w). And the loss function used in the training of the MLP deep neural network after introducing the physical model is as follows:

[0065]

[0066]

[0067] The above process is essentially a Lagrange multiplier method for solving the minimum value problem with the control equation as the constraint condition, and the detailed process is shown in the figure.

[0068] Therefore, the present application embeds the physical model constraint on the basis of the classical MLP flow field prediction model Figure 10 The flow control equation and the boundary condition constraint are considered to realize the high-precision complex flow field prediction. The flow field prediction result is verified in detail through relevant high-precision CFD calculation and wind tunnel test, so as to ensure the solution accuracy of the flow field rapid prediction model, and the detailed process is shown in the formula (2). Figure 11

[0069] S4: The three-dimensional flow field rapid prediction of the specified nozzle geometric feature parameters, flow state parameters and flow control parameters is realized through the flow field prediction model, and the flow field information of the specified nozzle is obtained.

[0070] ​Further, in an embodiment of the present application, after the three-dimensional flow field is quickly predicted by the flow field prediction model for the specified nozzle geometric feature parameters, flow state parameters and flow control parameters, the flow field information is obtained, and further comprising:

[0071] The aerodynamic characteristic parameters of the specified nozzle are obtained through the three-dimensional flow field information, including lift, thrust, core volume and thrust bias degree.

[0072] Further, in an embodiment of the present application, further comprising:

[0073] By changing the geometric feature parameters of the specified nozzle, the flow field prediction model is used to solve the nozzle aerodynamic characteristics under different flow state parameters and flow control parameters, so as to screen out sensitive parameters.

[0074] Further, in an embodiment of the present application, the three-dimensional flow field of the specified nozzle is quickly predicted by the flow field prediction model for the specified nozzle geometric feature parameters, flow state parameters and flow control parameters, and the three-dimensional flow field information of the specified nozzle is obtained, and further comprising:

[0075] The flow field pressure gradient is obtained through the three-dimensional flow field information, and the action mechanism of the sensitive parameters is analyzed through the wave distribution.

[0076] The present application fully combines the latest research progress of deep learning algorithm, builds a deep neural network, and constructs a nozzle geometry and flow field data set, so as to realize the extraction of nozzle geometric features, the high-precision and fast prediction of three-dimensional flow field distribution, and finally complete the sensitive parameter analysis of the nozzle, such as Figure 3 As shown in the figure. The above process mainly includes two modules, which are:

[0077] Module 1: fast flow field prediction based on deep learning;

[0078] Module 2: sensitive parameter analysis based on flow field wave distribution.

[0079] In order to realize the rapid prediction of the flow field distribution of the nozzle, the module one mainly includes two processing procedures: Step A, the geometric feature parameter extraction based on the AutoEncoder method; and Step B, the flow field rapid prediction based on the PI-MLP method. The former is the pretreatment module of the latter, the AutoEncoder deep neural network is used to extract the feature of the shape parameters and compresses them to 5-8 geometric feature parameters, in this process, the nozzle geometric data set containing different configurations is introduced. Then, the nozzles under different geometric configurations are combined with various flow state parameters (Mach number, flow angle, Reynolds number, etc.) and various flow control parameters as the design variables of the nozzle, and the high-precision CFD simulation under different working conditions is carried out to obtain the flow field training data set, which is used to train the PI-MLP flow field rapid prediction deep neural network built in Step B. When the neural network is trained, the three-dimensional flow field of the nozzle under the specified working condition can be rapidly predicted, and the aerodynamic characteristic parameters of the nozzle such as lift, thrust, core volume and thrust bias degree can be obtained through relevant post-processing. Therefore, the module one is the basis and core of the present application.

[0080] The module two is mainly based on the high-precision and rapid prediction ability of the module one: the PI-MLP flow field prediction deep neural network is used to rapidly and accurately solve the aerodynamic characteristics of the nozzle under different working conditions by changing the geometric feature parameters in a certain range, so as to screen out the sensitive parameters. According to the pressure gradient distribution of the nozzle, the influence law of the sensitive parameters such as geometric, flow state and flow control on the wave system distribution is intuitively analyzed, and the action mechanism of the sensitive parameters is explored.

[0081] The three-dimensional nozzle flow field rapid prediction and sensitive parameter analysis method provided by the embodiment of the present application fully combines the AutoEncoder and the multi-layer perceptron technology embedded with physical constraints to build a three-dimensional flow field rapid prediction deep neural network for the nozzle; the nozzle geometric data set is constructed through parameterized modeling; the flow field data set required for neural network training is constructed by writing the grid division software and the corresponding script file of the computational fluid dynamics solver to realize large-scale batch simulation; by changing the nozzle geometric parameters and flow control parameters in a certain range and giving the flow state parameters, the trained flow field rapid prediction deep neural network is called to obtain the corresponding flow field velocity, pressure and other parameter distributions in real time, and the characteristic physical quantities of the nozzle under various working conditions such as lift, thrust and thrust bias are obtained through post-processing, and then the sensitive parameters of the nozzle are obtained by using the statistical method; the action mechanism of the sensitive parameters is analyzed by calculating the flow field pressure gradient and intuitively observing the flow field wave system distribution. The solving idea provided by the present application has a wide application range, greatly improves the calculation efficiency of the three-dimensional flow field of the nozzle, ensures the same order of magnitude of the flow field prediction accuracy and the CFD simulation, and reduces the time required for the sensitivity analysis.

[0082] In order to realize the above-mentioned embodiment, the application further provides a three-dimensional nozzle flow field rapid prediction and sensitive parameter analysis method device.

[0083] Figure 2 A three-dimensional nozzle flow field rapid prediction and sensitive parameter analysis method device provided by the embodiment of the application has the structure as shown in the figure.

[0084] As Figure 2 shown, the three-dimensional nozzle flow field rapid prediction and sensitive parameter analysis method device comprises an extraction module 10, a construction module 20, a training module 30 and a prediction module 40, wherein the extraction module is used to obtain nozzle geometric shape parameters, perform feature extraction on the geometric shape parameters by using a self-encoder neural network, and obtain geometric feature parameters; the construction module is used to reconstruct the nozzle geometric shape parameters by using the geometric feature parameters, take nozzle flow state parameters and flow control parameters as design variables, perform CFD simulation, and obtain nozzle flow field training data sets; the training module is used to build a multilayer perceptron neural network, train the multilayer perceptron neural network by using the training data sets, and introduce a flow control equation into a loss function of the training to obtain a flow field prediction model; and the prediction module is used to realize three-dimensional flow field rapid prediction of specified nozzle geometric feature parameters, flow state parameters and flow control parameters by using the flow field prediction model, and obtain flow field information of the specified nozzle.

[0085] Further, in an embodiment of the application, an analysis module is further included, which is used to:

[0086] acquire aerodynamic characteristic parameters of the specified nozzle by using the three-dimensional flow field information, including lift, thrust, core volume and thrust bias degree;

[0087] change the geometric feature parameters of the specified nozzle, solve the nozzle aerodynamic characteristics under different flow state parameters and flow control parameters by using the flow field prediction model, and screen out sensitive parameters;

[0088] acquire flow field pressure gradients by using the three-dimensional flow field information, analyze the action mechanism of the sensitive parameters by using wave series distribution.

[0089] In order to realize the above-mentioned embodiment, the application further provides a computer device, characterized by comprising a memory, a processor and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the computer program to realize the three-dimensional nozzle flow field rapid prediction and sensitive parameter analysis method.

[0090] In order to realize the above-mentioned embodiments, the application further provides a computer readable storage medium, which stores a computer program, wherein the computer program is executed by a processor to realize the three-dimensional nozzle flow field fast prediction and sensitivity parameter analysis method.

[0091] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, different embodiments or examples described in the present specification and the features of different embodiments or examples can be combined and combined by those skilled in the art without contradiction.

[0092] In addition, the terms "first", "second" are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "a plurality of" is at least two, for example, two, three, etc., unless otherwise specifically limited.

[0093] Although the embodiments of the present application have been shown and described above, it can be understood that the above-mentioned embodiments are exemplary and cannot be understood as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above-mentioned embodiments within the scope of the present application.

Claims

1. A method for rapid prediction of flow field and sensitivity parameter analysis of a three-dimensional nozzle, characterized in that, The method comprises the following steps: obtaining nozzle geometry parameters, using an autoencoder neural network to extract features of the geometry parameters to obtain geometry feature parameters; reconstructing the nozzle geometry parameters through the geometry feature parameters, taking the nozzle flow state parameters and flow control parameters as design variables, and obtaining flow field training data sets of the nozzle through CFD simulation; building a multilayer perceptron neural network, training the multilayer perceptron neural network through the training data sets, and introducing a flow control equation into a loss function of the training to obtain a flow field prediction model; realizing fast prediction of a three-dimensional flow field of specified nozzle geometry feature parameters, flow state parameters and flow control parameters through the flow field prediction model to obtain flow field information of the specified nozzle.

2. The method of claim 1, wherein, The method further comprises the following steps in the feature extraction of the geometry parameters by using the autoencoder neural network to obtain the geometry feature parameters: adding random noise to the geometry parameters.

3. The method of claim 1, wherein, Before the CFD simulation, the method further comprises the following steps: setting grid division and boundary conditions.

4. The method of claim 1, wherein, After the fast prediction of the three-dimensional flow field of the specified nozzle geometry feature parameters, flow state parameters and flow control parameters through the flow field prediction model to obtain the flow field information, the method further comprises the following steps: obtaining aerodynamic characteristic parameters of the specified nozzle through the three-dimensional flow field information, including lift, thrust, core volume and thrust bias degree.

5. The method of claim 4, wherein, The method further comprises the following steps: changing the geometry feature parameters of the specified nozzle, solving the nozzle aerodynamic characteristics under different flow state parameters and flow control parameters by using the flow field prediction model, and screening sensitive parameters.

6. The method according to claim 1 or 5, characterized in that, The method further comprises the following steps in the fast prediction of the three-dimensional flow field of the specified nozzle geometry feature parameters, flow state parameters and flow control parameters through the flow field prediction model to obtain the three-dimensional flow field information of the specified nozzle: obtaining flow field pressure gradients through the three-dimensional flow field information, analyzing the mechanism of sensitive parameters through wave distribution.

7. A device for rapid prediction of flow field and sensitivity parameter analysis of a three-dimensional nozzle, characterized in that, The method comprises the following steps: a extraction module for obtaining nozzle geometry parameters, using an autoencoder neural network to extract features of the geometry parameters to obtain geometry feature parameters; a construction module for reconstructing the nozzle geometry parameters through the geometry feature parameters, taking the nozzle flow state parameters and flow control parameters as design variables, and obtaining flow field training data sets of the nozzle through CFD simulation; a training module for building a multilayer perceptron neural network, training the multilayer perceptron neural network through the training data sets, and introducing a flow control equation into a loss function of the training to obtain a flow field prediction model; a prediction module for realizing fast prediction of a three-dimensional flow field of specified nozzle geometry feature parameters, flow state parameters and flow control parameters through the flow field prediction model to obtain flow field information of the specified nozzle.

8. The apparatus of claim 7, wherein, The method further comprises the following steps in the fast prediction of the three-dimensional flow field of the specified nozzle geometry feature parameters, flow state parameters and flow control parameters through the flow field prediction model to obtain the three-dimensional flow field information of the specified nozzle: obtaining aerodynamic characteristic parameters of the specified nozzle through the three-dimensional flow field information, including lift, thrust, core volume and thrust bias degree. By changing the geometric parameters of the specified nozzle, the flow field prediction model is used to solve the nozzle aerodynamic characteristics under different flow state parameters and flow control parameters to screen out sensitive parameters; The three-dimensional flow field information is used to obtain flow field pressure gradient, and the mechanism of sensitive parameters is analyzed through wave distribution.

9. A computer device, comprising: The computer program is stored in the memory and executable on the processor, and the processor executes the computer program to realize the three-dimensional nozzle flow field fast prediction and sensitive parameter analysis method of any one of claims 1-6.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the three-dimensional nozzle flow field fast prediction and sensitive parameter analysis method of any one of claims 1-6.

Images